Comprehensive Product Overview

The Mellanox MQM9790-HS2F (NVIDIA Quantum-2 QM9790) is an unmanaged, high-radix InfiniBand switch purpose-built for 400Gb/s NDR environments. Housed in a compact 1U rack-mount chassis, it provides 64 non-blocking NDR 400Gb/s InfiniBand ports via 32 OSFP (Octal SFP) connectors—each OSFP supports two logical 400Gb/s ports.

Powered by the NVIDIA Quantum-2 ASIC and an x86 Coffee Lake i3 CPU with 8GB DDR4 memory, the MQM9790-HS2F is an unmanaged model requiring an external Subnet Manager (SM) for fabric management. It features port-to-cable (P2C/front-to-rear) airflow, 1+1 redundant hot-swap 2000W AC power supplies, and N+1 hot-swap fans for maximum reliability and uptime.

With a switching capacity of 51.2Tb/s and a forwarding rate of 66.5B+ packets per second (BPPS), it delivers wire-speed performance across all ports with ultra-low latency for AI and HPC workloads. The switch supports SHARP v3 for in-network computing, accelerating collective operations by up to 32× and reducing data movement. Backward compatibility with HDR/EDR/FDR/QDR InfiniBand (200G/100G/56G/40G) ensures seamless integration with legacy infrastructures.

Key Features & Business Benefits

Key Features

• 64×400Gb/s NDR InfiniBand Ports: 32 OSFP connectors; 51.2Tb/s non-blocking throughput。
• Quantum-2 ASIC + SHARP v3: 3rd-gen in-network computing; 32× faster collective operations。
• Unmanaged (External SM): Optimized for large fabrics with centralized management。
• 1U Rack-Mount, P2C Airflow: 1.7”×17.2”×26”; front-to-rear cooling。
• Redundant Hot-Swap Components: 1+1 2000W AC PSUs; N+1 hot-swap fans。
• x86 Coffee Lake i3 CPU: 8GB DDR4; MLNX-OS with CLI/WebUI/SNMP。
• Ultra-Low Latency: <100ns port-to-port (NDR InfiniBand)。
• Adaptive Routing & Congestion Control: Maximizes fabric efficiency。
• Backward Compatibility: Supports HDR/EDR/FDR/QDR InfiniBand。

Business Benefits

• Accelerate AI/HPC Workloads: 400Gb/s bandwidth + SHARP v3 cuts AI training time by 50%+。
• Maximize Data Center Density: 64×400G ports in 1U reduces rack space by 70% vs. legacy switches。
• Optimize Large-Scale Fabric Management: Unmanaged design scales to 10,000+ nodes with external SM。
• Ensure 99.999% Uptime: Redundant hot-swap PSUs/fans minimize downtime。
• Reduce TCO: 80+ Platinum-certified PSUs lower power consumption by 40%。
• Future-Proof Investment: NDR 400G ready; backward compatible with older InfiniBand generations。

Technical Specifications

Ideal Application Scenarios

• AI/ML Training Clusters: Large-scale GPU clusters (e.g., NVIDIA DGX) for deep learning & LLMs。
• High-Performance Computing (HPC): Supercomputing, scientific simulation, CFD, FEA, & quantum computing。
• Enterprise Data Centers: High-density virtualization, NVMe-oF storage, & big data analytics。
• Cloud & Hyperscale: Scalable InfiniBand fabrics for cloud HPC/AI services。
• Research Labs: University supercomputing centers & government research clusters。

Supported Accessories & Modules

• OSFP Optics: 400Gb/s NDR InfiniBand transceivers (SR4, LR4, ER4, DR8)。
• Breakout Cables: 400Gb/s-to-2×200Gb/s HDR passive/active copper cables。
• Power Supplies: MTEF-PSF-AC-2000W (P2C airflow, hot-swap)。
• Fan Modules: MTEF-FANF-P2C (P2C airflow, hot-swap)。
• Rack Accessories: Rail kit (included), cable management arms, 1U blanking panels。
• Management Tools: NVIDIA MLNX-OS, OFED drivers, centralized SM software。

Frequently Asked Questions (FAQ)

Q1: What is the difference between MQM9790-HS2F and MQM9700-HS2F?

A: MQM9790-HS2F = unmanaged (external SM); MQM9700-HS2F = managed (onboard SM)。

Q2: Can it work with HDR/EDR InfiniBand hardware?

A: Yes—backward compatible with HDR (200G), EDR (100G), FDR (56G), & QDR (40G)。

Q3: What is the maximum cluster size supported?

A: Up to 10,000 nodes with an external Subnet Manager。

Q4: Does it support remote management?

A: Yes—MLNX-OS provides CLI, WebUI, SNMP, & JSON APIs for remote monitoring/configuration。

Q5: What is the typical power consumption?

A: ~747W (passive cables); ~1,720W (active cables, max load)